aboutsummaryrefslogtreecommitdiffstats
path: root/extra/cleanups/file_meta.md
blob: d10d0e0fb8565de8e47d1a3fd48dbb95cccd9eca (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227

Over 500k file entities still lack complete metadata. For example, SHA-256
checksums and verified mimetypes.

Presumably these also lack GROBID processing. It seems that most or all of
these are simply wayback captures with no CDX metadata in sandcrawler-db, so
they didn't get update in prior cleanups.

Current plan, re-using existing tools and processes, is to:

1. create stub ingest requests containing file idents
2. process them "locally" on a large VM, in 'bulk' mode; writing output to stdout but using regular grobid and pdfextract "sinks" to Kafka
3. transform ingest results to a form for existing `file_meta` importer
4. run imports

The `file_meta` importer requires just the `file_meta` dict from sandcrawler.

## Prep

    zcat file_hashes.tsv.gz | pv -l | rg '\t\t' | wc -l
    # 521,553

    zcat file_export.json.gz \
        | rg -v '"sha256":' \
        | pv -l \
        | pigz \
        > files_missing_sha256.json.gz
    # 521k 0:10:21 [ 839 /s]

Want ingest requests with:

    base_url: str
    ingest_type: "pdf"
    link_source: "fatcat"
    link_source_id: file ident (with "file_" prefix)
    ingest_request_source: "file-backfill"
    ext_ids:
        sha1: str

Use `file2ingestrequest.py` helper:

    zcat files_missing_sha256.json.gz \
        | ./file2ingestrequest.py \
        | pv -l \
        | pigz \
        > files_missing_sha256.ingest_request.json.gz
    # 519k 0:00:19 [26.5k/s]

So about 2k filtered out, will investigate later.

    zcat files_missing_sha256.ingest_request.json.gz \
        | shuf -n1000 \
        > files_missing_sha256.ingest_request.sample.json

    head -n100 files_missing_sha256.ingest_request.sample.json | ./ingest_tool.py requests --no-spn2 - > sample_results.json
         4 "no-capture"
         1 "no-pdf-link"
        95 "success"

Seems like this is going to be a good start, but will need iteration.

Dev testing:

    head files_missing_sha256.ingest_request.sample.json \
        | ./ingest_tool.py file-requests-backfill - --kafka-env qa --kafka-hosts wbgrp-svc263.us.archive.org:9092,wbgrp-svc284.us.archive.org:9092,wbgrp-svc285.us.archive.org:9092 \
        > out_sample.json


## Commands

Production warm-up:

    cat /srv/sandcrawler/tasks/files_missing_sha256.ingest_request.sample.json \
        | ./ingest_tool.py file-requests-backfill - --kafka-env prod --kafka-hosts wbgrp-svc263.us.archive.org:9092,wbgrp-svc284.us.archive.org:9092,wbgrp-svc285.us.archive.org:9092 --grobid-host http://localhost:8070 \
        > /srv/sandcrawler/tasks/files_missing_sha256.ingest_results.sample.json

Production parallel run:

    zcat /srv/sandcrawler/tasks/files_missing_sha256.ingest_request.json \
        | parallel -j24 --linebuffer --round-robin --pipe ./ingest_tool.py file-requests-backfill - --kafka-env qa --kafka-hosts wbgrp-svc263.us.archive.org:9092,wbgrp-svc284.us.archive.org:9092,wbgrp-svc285.us.archive.org:9092 --grobid-host http://localhost:8070 \
        > /srv/sandcrawler/tasks/files_missing_sha256.ingest_results.json

Filter and select file meta for import:

    head files_missing_sha256.ingest_results.json \
        | rg '"sha256hex"' \
        | jq 'select(.request.ext_ids.sha1 == .file_meta.sha1hex) | .file_meta' -c \
        > files_missing_sha256.file_meta.json
    # Worker: Counter({'total': 20925, 'success': 20003, 'no-capture': 545, 'link-loop': 115, 'wrong-mimetype': 104, 'redirect-loop': 46, 'wayback-error': 25, 'null-body': 20, 'no-pdf-link': 18, 'skip-url-blocklist': 17, 'terminal-bad-status': 16, 'cdx-error': 9, 'wayback-content-error': 4, 'blocked-cookie': 3})
    # [etc]


Had some GROBID issues, so are not going to be able to get everything in first
pass. Merge our partial results, as just `file_meta`:

    cat files_missing_sha256.ingest_results.batch1.json files_missing_sha256.ingest_results.json \
        | jq .file_meta -c \
        | rg '"sha256hex"' \
        | pv -l \
        > files_missing_sha256.file_meta.json
    # 386k 0:00:41 [9.34k/s]

A bunch of these will need to be re-run once GROBID is in a healthier place.

Check that we don't have (many) dupes:

    cat files_missing_sha256.file_meta.json \
        | jq .sha1hex -r \
        | sort \
        | uniq -D \
        | wc -l
    # 86520

Huh, seems like a weirdly large number. Maybe related to re-crawling? Will need
to dedupe by sha1hex.

Check how many dupes in original:

    zcat files_missing_sha256.ingest_request.json.gz | jq .ext_ids.sha1 -r | sort | uniq -D | wc -l

That lines up with dupes expected before SHA-1 de-dupe run.

    cat files_missing_sha256.file_meta.json \
        | sort -u -S 4G \
        | pv -l \
        > files_missing_sha256.file_meta.uniq.json

    cat files_missing_sha256.file_meta.uniq.json \
        | jq .sha1hex -r \
        | sort \
        | uniq -D \
        | wc -l
    # 0

Have seen a lot of errors like:

    %4|1637808915.562|TERMINATE|rdkafka#producer-1| [thrd:app]: Producer terminating with 1 message (650 bytes) still in queue or transit: use flush() to wait for outstanding message delivery

TODO: add manual `finish()` calls on sinks in tool `run` function

## QA Testing

    export FATCAT_API_AUTH_TOKEN... # sandcrawler-bot

    cat /srv/fatcat/datasets/files_missing_sha256.file_meta.uniq.sample.json \
        | ./fatcat_import.py --editgroup-description-override 'backfill of full file-level metadata for early-imported papers' file-meta -
    # Counter({'total': 1000, 'update': 503, 'skip-existing-complete': 403, 'skip-no-match': 94, 'skip': 0, 'insert': 0, 'exists': 0})

    head -n1000 /srv/fatcat/datasets/files_missing_sha256.file_meta.uniq.json \
        | parallel -j8 --round-robin --pipe -q ./fatcat_import.py --editgroup-description-override 'backfill of full file-level metadata for early-imported papers' file-meta -
    # Counter({'total': 1000, 'update': 481, 'skip-existing-complete': 415, 'skip-no-match': 104, 'skip': 0, 'insert': 0, 'exists': 0})

## 2021-11-29 Update

    zcat ../2021-11-25/file_export.json.gz \
        | rg -v '"sha256":' \
        | pv -l \
        | gzip \
        > files_missing_sha256.json.gz
    # 356k 0:11:13 [ 529 /s]

As a side note, almost all the missing entities are at the "start" of the
export file, not the "end". Presumably because they were imported early on?

    head result_cdx.json | ./pdfextract_tool.py -j1 extract-json -

The `pdfextract_tool.py` already does Kafka publishing, which is great. Should
be much faster than GROBID worker; we can do the GROBID re-processing later.

Use new sandcrawler CDX lookup/fetch script to find exact CDX rows:

    # in sandcrawler/python folder
    zcat /schnell/fatcat_cleanups/file_meta/files_missing_sha256.json.gz \
        | parallel -j6 --round-robin --pipe --linebuffer pipenv run python -m scripts.fetch_cdx_sha1hex - \
        | pv -l \
        | gzip \
        > /schnell/fatcat_cleanups/file_meta/files_missing_sha256.fetched.json.gz
    # 356k 2:17:27 [43.2 /s]

    # stats
    zcat /schnell/fatcat_cleanups/file_meta/files_missing_sha256.fetched.json.gz \
        | jq .status -r \
        | sort \
        | uniq -c

      15477 fail-not-found
          5 skip-no-urls
     314291 success-api
      26723 success-db

    # extract all CDX rows for processing
    zcat /schnell/fatcat_cleanups/file_meta/files_missing_sha256.fetched.json.gz \
        | rg '"success' \
        | jq .cdx_rows[] -c \
        | pv -l \
        | shuf \
        | gzip \
        > /schnell/fatcat_cleanups/file_meta/files_missing_sha256.cdx_rows.json.gz
    # 354k 0:00:17 [19.7k/s]

    export TMPDIR=/fast/tmp/
    zcat /schnell/fatcat_cleanups/file_meta/files_missing_sha256.cdx_rows.json.gz \
        | parallel -j16 --round-robin --pipe --linebuffer pipenv run ./pdfextract_tool.py -j1 extract-json - \
        | pv -l \
        | gzip \
        > /schnell/fatcat_cleanups/file_meta/files_missing_sha256.pdf_extract.json.gz
    # 354k 2:45:35 [35.7 /s]

    zcat /schnell/fatcat_cleanups/file_meta/files_missing_sha256.pdf_extract.json.gz | jq .status -r | sort | uniq -c | sort -nr

     299226 success
      44297 parse-error
       8986 error-wayback
       1819 not-pdf
        518 text-too-large
          1 empty-blob

    zcat /schnell/fatcat_cleanups/file_meta/files_missing_sha256.pdf_extract.json.gz \
        | jq .file_meta -c \
        | rg -v ^null \
        | pv -l \
        | pigz \
        > /schnell/fatcat_cleanups/file_meta/files_missing_sha256.file_meta.json.gz
    # 345k 0:08:21 [ 689 /s]

Holding off on actually importing, because metadata dumps are happening on
fatcat prod database server.